Marco-o1 // Open Reasoning Models for Open-Ended Solutions
OpenAI o1 model started a new wave of “reasoning” models // now open models
Inference with ollama
# install ollama
!curl -fsSL https://ollama.com/install.sh | sh
# start ollama
!nohup ollama serve > ollama.out 2>&1 &
# pull marco-o1
!ollama pull marco-o1Test marco-o1
import ollama
model='marco-o1'
response = ollama.chat(model=model, messages=[
{
'role': 'user',
'content': 'why sky is blue?',
},
])
print(response['message']['content'])
<Thought>
Alright, I need to explain why the sky appears blue. Let’s start by recalling some basic physics principles related to light and how it interacts with Earth’s atmosphere.
First, sunlight enters our atmosphere composed of various colors, which we perceive as white light. This light consists of different wavelengths, each corresponding to a specific color in the visible spectrum. Blue light has shorter wavelengths compared to other colors like red or orange, which have longer wavelengths.
When sunlight encounters the Earth’s atmosphere, it collides with gas molecules and tiny particles in the air. These collisions cause the light to scatter in all directions. The principle governing this scattering is known as Rayleigh scattering, named after Lord Rayleigh who first described it mathematically.
According to Rayleigh’s theory, shorter wavelengths of light are scattered more effectively than longer ones. Since blue light has a shorter wavelength, it gets scattered more efficiently by the air molecules in the atmosphere. This means that when we look up at the sky during clear conditions, most of the blue light from the sun is being scattered towards our eyes from all directions.
But wait, why doesn’t the sky appear violet then? That’s another interesting point. Violet light has even shorter wavelengths than blue, so it should scatter more. However, human eyes are less sensitive to violet light compared to blue. Additionally, there’s a higher concentration of red and orange wavelengths in sunlight that get absorbed by air molecules, further reducing the overall amount of scattered violet light we perceive.
Another factor to consider is the time of day. During sunrise and sunset, the sun is lower in the sky, and its light has to travel through more of Earth’s atmosphere. This increased path length causes even more scattering of blue and green light, leaving predominantly red and orange hues visible at those times. That’s why the sky often appears red or orange during these periods.
Moreover, atmospheric conditions like pollution can affect the scattering process. Particles in the air can scatter different wavelengths of light differently, potentially altering the color perception of the sky. For example, high levels of pollutants might cause more yellow or even white hues to dominate the sky’s appearance.
It’s also worth noting that on planets with atmospheres composed of different gases or with varying concentrations of particles, the sky colors would differ. Mars, for instance, has a pinkish sky due to its thin atmosphere and dust particles.
In summary, the blue color of the sky is primarily due to Rayleigh scattering of shorter-wavelength light by Earth’s atmospheric molecules, combined with human visual sensitivity differences compared to other colors like violet.
</Thought>
<Output>
The sky appears blue because sunlight enters our atmosphere and interacts with gas molecules. Blue light has shorter wavelengths than other colors in the visible spectrum. When sunlight collides with these molecules, it scatters in all directions through a process called Rayleigh scattering. Since blue light is scattered more effectively due to its shorter wavelength, it dominates the sky’s color when viewed from Earth during clear conditions.
</Output>
Marco-o1 not only focuses on disciplines with standard answers, such as mathematics, physics, and coding — which are well-suited for reinforcement learning (RL) — but also places greater emphasis on open-ended resolutions. We aim to address the question: “Can the o1 model effectively generalize to broader domains where clear standards are absent and rewards are challenging to quantify?”
Currently, Marco-o1 Large Language Model (LLM) is powered by Chain-of-Thought (CoT) fine-tuning, Monte Carlo Tree Search (MCTS), reflection mechanisms, and innovative reasoning strategies — optimized for complex real-world problem-solving tasks.
